Skip to main content
All terms
Training

Test-Time Scaling

Improving results by spending more compute while answering, instead of only by training bigger models.

Definition

Test-time scaling is the idea of getting better answers by spending extra computation at the moment of use — through longer reasoning chains, multiple tries, or search — rather than only by training a larger model. It marks a shift from 'scale the model' toward 'scale the thinking' at the point of the question.